|
In January, the Full Court Press introduced its first experiment with Claude Code — a case tracker that catches recent ONCA and SCC decisions and ranks them by buzz. That was the starting point. Since then, the experimentation has continued, and the toolkit has grown. Here’s an honest update on what’s been built, what works, and what doesn’t. Beyond the Chatbot Before getting into the tools, it’s worth explaining what’s actually happening here — because it’s not what most people picture when they hear “AI.” Most people’s experience with artificial intelligence starts and ends with a chatbot. You type a question into ChatGPT, it types back an answer, and that’s the interaction. It’s impressive for what it is, but it’s also limited. A chatbot is a conversationalist. It talks. That’s it. What the Full Court Press has been experimenting with is something different: agentic AI. Instead of just answering questions, an AI agent can take actions. It can search databases, read documents, move files, write code, call APIs, and chain those steps together to complete a task from start to finish. The human provides the goal; the agent figures out the steps. Think of it this way: a chatbot is like asking a colleague a question over coffee. An agentic tool is like handing a task to a junior associate and getting back a completed work product. The associate doesn’t just tell you what the law might be — they go to CanLII, pull the cases, read the decisions, and draft the memo. That’s the difference. This distinction matters because the public conversation about AI in law is still largely stuck on chatbots. Lawyers either dismiss AI because ChatGPT hallucinated a case citation (fair criticism), or they embrace it as a fancy search engine (underselling it). The reality is that AI tools have moved well past the “ask a question, get an answer” paradigm. The tools described below don’t just talk — they do things. They run on schedules, interact with real systems, and produce tangible outputs. Understanding that shift is key to understanding where legal technology is actually heading. The Case Tracker (Update): The original experiment is still running. It pulls recent criminal decisions from the Ontario Court of Appeal and the Supreme Court of Canada, summarizes them, and ranks them by how much attention they’re getting online. It remains a useful daily check-in tool. No major changes — it does what it was designed to do. The Calendar & Client Manager: Next came a password-protected calendar and task management tool, paired with a basic client tracker. The idea was simple: one place to see upcoming dates, deadlines, and client information without toggling between apps. It works. It’s not fancy. It does the job of keeping matters organized and accessible from any device. The client management side lets you store key details for each file and pull them up quickly. For a solo or small practice, it removes friction. The Downloads Sorter: This one came out of a very specific frustration. Criminal defence lawyers deal with a constant stream of documents — disclosure, court orders, bail packages, correspondence — and they all land in the same Downloads folder. The Downloads Sorter automatically moves files into the correct client folder on Google Drive. It works by matching filenames and document content against client names. It runs daily on a schedule, quietly sorting files overnight. When it can’t confidently match a file, it sends it to an “Unsorted” folder rather than guessing wrong. The difficulties here were real. Early versions used simple text matching, which led to false positives — a client named “Ola” was matching every file containing “Fashola,” and legal documents that cite dozens of case names were getting matched to the wrong clients entirely. The fix was switching to strict word-boundary matching and teaching the system learned associations (for example, telling it that co-accused files should go to the same client folder). It’s now reliably sorting files with very few errors. The honest assessment: it saves 10–15 minutes of manual filing per day, which adds up. The Legal Memo Maker: This is the most ambitious tool and the one with the most potential — and the most problems. The idea: you type a legal question, and the tool searches CanLII using a free API, finds and ranks relevant cases, reads the actual decision text, and drafts an IRAC-style legal memo with real citations and real quotes. Why it matters: the biggest criticism of AI in legal research is hallucination. Language models fabricate case names, invent quotes, and cite decisions that don’t exist. This tool was designed to solve that problem by forcing the AI to read the actual decisions before writing — no guessing, no fabricating. Every citation and every quote must come from text the system has actually retrieved and processed. The difficulties have been significant. CanLII doesn’t provide full decision text through its API — only metadata. So the tool has to visit the actual case pages and extract the text. It’s fragile. The honest assessment: when it works, it’s genuinely impressive. The memos cite real cases, pull real quotes, and follow proper legal analysis structure. When the scraping fails, the tool falls back to metadata-only memos, which are less useful. This tool is not reliable enough for daily dependence, but it’s a promising proof of concept. The core insight — that AI-drafted legal writing must be grounded in actual source text — is the right approach, even if the plumbing still needs work. Overall Reflections -- A few honest takeaways after two months of building: 1. The simple tools are the most useful. The case tracker and downloads sorter run quietly in the background and save real time every day. They don’t try to do too much. 2. AI is better at organizing than it is at thinking. Sorting files, pulling case metadata, ranking search results — these are tasks where AI adds clear value. Drafting legal analysis is harder and requires much more guardrailing to prevent the AI from confidently making things up. 3. Agentic tools are where the real value is. The leap from chatbot to agent is the leap from “that’s a neat trick” to “that just saved me an hour.” The tools that do things — move files, search databases, draft documents from real sources — are fundamentally more useful than the ones that just answer questions. 4. Nothing is finished. Every tool described here has broken at least once and required fixing. This is not plug-and-play software. It’s experimental, iterative, and held together with a willingness to troubleshoot. That said — each tool is more stable today than it was a month ago. 5. The gap between “possible” and “reliable” is where all the work lives. Getting a prototype running takes hours. Getting it to work consistently takes weeks. Most of the effort described above wasn’t building features — it was fixing edge cases, handling errors, and dealing with systems that weren’t designed to cooperate with each other. The Full Court Press will keep experimenting. The goal remains the same: small, practical tools that help criminal defence lawyers stay efficient and informed. Not everything will work. But the ones that do are already saving time. AND YES: This post was written by Claude Code.
0 Comments
To start 2026, the Full Court Press plans on experimenting with another AI developer tool called "Claude Code". Think "coding for non-coders". The Full Court Press will aim to design and implement small automations to assist criminal defence lawyers stay efficient and informed. Note: This is all experimental, iterative, and a "work in progress". Use with caution! First experiment: "SCC and ONCA Case Catcher!" I asked Claude Code to design a website / app that collects recent ONCA and SCC cases (from the last 30 and 90 days respectfully) and provide summaries (with an option to have them read to the user out loud). The "caught" cases are then arranged / ranked according to the amount of "buzz" they are getting online. Check it out! https://tracker.natejackson.ca/
In this episode, Nate breaks down R. v. Ahmadi, a 2025 decision from the Ontario Court of Appeal, focusing on a key evidentiary issue: when an accused’s out-of-court statements can be used against them.
The Court draws a sharp line between disbelieving a statement and finding it was fabricated. Only fabricated statements—with independent evidence backing that up—can be used as circumstantial evidence of guilt. We talk through how Ahmadi’s ever-changing police interview crossed that line, how the trial judge handled it, and why the ONCA upheld the conviction. A must-know case for anyone running—or challenging—post-offence statements. Plus, a lawyer joke to send you off smiling. https://rss.com/podcasts/the-full-court-press/1957751/ Title: Presumed Innocent: The Supreme Court’s Ruling on Prison Discipline
Length: ~5 minutes In this episode, Nate and ChatGPT break down the Supreme Court of Canada’s landmark decision in John Howard Society of Saskatchewan v. Saskatchewan (2025 SCC 6). The Court ruled that serious prison disciplinary proceedings—those that can result in segregation or loss of earned remission—must meet the criminal standard of proof beyond a reasonable doubt. rss.com/podcasts/the-full-court-press/1957445/ In this episode, criminal defense lawyer Nate Jackson and co-host ChatGPT break down two fresh decisions from the Ontario Court of Appeal — R. v. Arora and R. v. James.
We talk fake guns, real trauma, and whether pleading guilty still buys you anything (Arora), then shift gears to police overreach, Charter breaches, and the cost of ignoring the rules (James). Quick, punchy, and packed with insights — plus a closing legal joke, because why not? Tune in. Court is in session. https://rss.com/podcasts/the-full-court-press/1957501/ Hello everyone!
I'm thrilled to announce that I'll be experimenting with AI models here at The Full Court Press to dive deep into criminal law topics. This new venture aims to bring a fresh, dynamic approach to discussing complex legal issues. I'll be using AI to provide insights, answer questions, and offer perspectives that can help simplify and clarify criminal law concepts. Stay tuned as we explore how AI can help break down legal jargon and provide a clearer understanding of the law! I’m excited to see where this experiment leads. Cheers, Nate (and yes, this post was created via ChatGPT... I'm not actually "thrilled" to announce this. I'm intruiged more than anything as an AI skeptic but, for better or for worse, a curious AI skeptic. Let's see what happens...) The takeaway: A court-appointed amicus curiae lawyer has a limited, but important, role in cases in which they’re assigned. They must uphold a duty of loyalty to the court and act as a lawyer of, and for, the court. Their functions cannot go against the duty of impartiality. The big picture:
In a case for two counts of frst-degree murder, Mr. Kahsai sought to represent himself. He regularly disrupted proceedings and failed to meaningfully advance a defence for himself or help move the matter along. The trial judge thus determined that it was essential to appoint amicus curiae in order to ensure Mr. Kahsai’s fair trial. This lawyer was appointed to cross-examine witnesses, though he was instructed not to advocate for Mr. Kahsai. Mr. Kahsai continued to be disruptive and resisted the court-appointed lawyer, deciding to be uncooperative. After Mr. Kahsai was convicted of the charges, he appealed on a number of grounds, including an unfairness brought upon by the trial judge’s failure to appoint amicus with an adversarial role earlier on in the trial. These appeals were dismissed. The Supreme Court held here that the trial judge has the discretion, broad in scope, to appoint amicus with specifc powers that suit the case at hand. While it may be open to the trial judge to appoint a more partial amicus, the trial judge should consider several factors in determining the scope and the partiality of the amicus. These factors include looking at the trial in its whole context: the nature and complexity of the charges, whether it is a judge-alone or jury trial, the attributes of the accused, what the assistance is needed for (whether that be to test the Crown’s case or advance a defence), and what assistance the Crown and trial judge can provide. Ahead of an appointment of amicus, the trial judge should canvass with the parties about the necessity of the appointment and the potential scope. The role of the amicus can be restricted by the rights of the accused to conduct their own defence (though this constitutional right is not absolute). Amicus appointments can also be restricted by other factors: They cannot be used to circumvent the inability to secure counsel at legal aid rates and the role cannot be given scope that conficts with the amicus role of loyalty to the court. Ultimately, in this case, the Supreme Court held that the discretionary decision made by the trial judge was appropriate in the dificult circumstances of a strenuous trial. Given that the accused had insisted upon advancing his own defence, the trial judge appointed an amicus with a limited role to help move the proceedings along. There was no irregularity found here to advance a finding of a miscarriage of justice, and the appeal was thus dismissed. Blog Post By: Gabe Nisker
The takeaway: Courts need to be careful about the use of lay opinion evidence. The conclusion: In this Ontario Court of Appeal decision, the court set aside convictions and ordered a new trial for Mr. Jenkins, who had previously been convicted on several charges pertaining to drug traficking. The full story: In the appeal, Mr. Jenkins’s counsel put forward that fve police oficers gave what’s called “lay opinion evidence” that was improperly admitted in the initial trial. Lay opinion evidence is opinion evidence not given by an expert, and it is only permissible under a limited number of circumstances outlined in the case of Graat v. The Queen. Here, the opinion evidence in question was that these oficers found Mr. Jenkins’s conduct to be consistent with drug traficking. The question before the court was whether these opinions could be permissible. In each instance, the oficer had been asked a question like “What do you make of this behaviour in your experience?” and ofered his opinion thereafter. Justice Copeland, relying on a previous case (R. v. Nguyen, 2023 ONCA 531), pointed to the various issues with this evidence. First, the Crown had not sought to qualify these surveillance oficers as oficial experts on indicators of drug traficking. Next, the jury could weigh the interactions as described by the officers without their additional conclusive opinions. Without those directly stated opinions, the officers could still clearly convey what they had seen, and the jury could still understand what was put before them. However, to distinguish this case from Nguyen and others, Justice Copeland declined to apply the curative proviso and ordered a new trial. The curative proviso would allow for the admission of the opinion evidence anyway, if the Crown can meet the burden of showing a minor or harmless error or if they show that the case was so overwhelming against the person appealing their conviction (here, Mr. Jenkins). Justice Copeland held that the signifcance of having this trial in front of a jury without the trial judge’s caution about the evidence showed great potential for misused evidence worthy of a new trial. Weighing the whole of the error, Justice Copeland saw that it was not minor or harmless. Further, to the use in the instance of an overwhelming case against Mr. Jenkins, Justice Copeland found that the number of live issues, including questions of credibility, meant that there was a strong, but not overwhelming, case. For that reason, the appeal was allowed, the convictions set aside, and a new trial was ordered. By: Gabe Nisker
The takeaway: Know your rights to be tried in the oficial language of your choice, whether that be English or French. First instance judges must make the accused aware of their Section 530(3) rights. The full story: In this case, Mr. Tompouba, a bilingual Francophone, brought an appeal on the grounds that there was an error in the application of the law. He claimed that he would have preferred to have his trial in French and that he did not receive the free and informed choice he should have under Section 530(3) of the Criminal Code. Section 530(3) enshrines an accused’s right to be advised of their fundamental right to be tried in the language of their choice (English or French) and how it is to be exercised properly. This section of the Criminal Code puts upon a judge a two-pronged duty, requiring them to ensure the accused is properly informed. They must take steps to make the accused aware of their rights and make them aware of how to exercise them. A frst appearance judge who fails to make the accused aware infringes this right. A breach of this right is considered an error of law under s. 686(1)(a)(ii) of the Criminal Code: If the frst appearance judge doesn’t make the accused aware of their right, this is an improper application of a legal rule, which gets in the way of the trial court’s proper judgment. The Crown can then relieve their burden using a “curative proviso,” which allows that the Court can dismiss any appeal if the Crown can show that no prejudice was caused by the error or irregularity. Accordingly, in this case, the frst instance judge failed to make Mr. Tompouba aware of his rights. Then, in the appeal, the Court of Appeal judge improperly placed the burden of proving the infringement of his rights on Mr. Tompouba. The Supreme Court re-emphasized that the curative proviso places the burden on the Crown to show no prejudice. The Supreme Court of Canada held that it ultimately cannot be concluded on a balance of probabilities that Mr. Tompouba would have elected to conduct his trial in English had he had the choice, and they granted his appeal, ordering a new French trial. The Court found that the Crown had not relieved their burden and that there may have been some prejudice caused. The Case: R. v. Kruk, 2024 SCC 7 In short: Appellate courts should adhere to established legal principles when assessing a trial judge's credibility or reliability assessments, considering whether alleged assumptions are indeed findings based on evidence, and applying the appropriate standard of review, such as correctness, for errors of law like reliance on myths, stereotypes, or biases. The Bottom Line: Per Wagner C.J. and Côté, Martin, Kasirer, Jamal and O’Bonsawin JJ: The rejection of a proposed rule against ungrounded common-sense assumptions underscores its divergence from established legal principles, particularly in sexual assault cases, where myths and stereotypes warrant distinct treatment. This rule, misaligned with the protective framework for sexual assault complainants, would extend questioning into irrelevant sexual history, potentially reviving prejudice. It contradicts appellate deference to trial judges and risks appellate overreach into credibility assessments. Appellate courts should adhere to existing standards, distinguishing palpable errors from reasonable assumptions, ensuring fairness without undermining trial judge expertise. Errors of law, such as reliance on stereotypes, warrant correction, but unsubstantiated assumptions demand deference, safeguarding the integrity of trial proceedings. Practice Point: At trial, take a coherent position regarding the difference between facts and assumptions. Consider using Rowe's proposed framework to protect the record on appeal: (1) What generalized expectation can the Judge rely on in their reasoning process? (2) Outline why it is (or isn't) reasonable to rely on this particular general expectation in the unique circumstances of this case? (3) Be clear that the Trial Judge cannot rely on the generalized expectation as itself a conclusive and indisputable fact. |
